|Ph.D Student||Ishaq Robert|
|Subject||Development of a Flexible Model Structure for Discrete|
|Department||Department of Civil and Environmental Engineering||Supervisors||Professor Yoram Shiftan|
|Professor Shlomo Bekhor|
|Full Thesis text|
Discrete choice models are widely used in the transportation planning field; however there is a major disadvantage in the way they are currently being implemented.
The estimation process usually uses the entire study population, and therefore leads to average estimated parameter values and average elasticities. Moreover, in a hierarchical model structure such as in an Activity Based Model, the researcher assumes a single-choice hierarchical order even though there are a number of possible options.
The segmentation processes in the field of travel demand primarily focus on dividing the data a priori according to one or more variables, based on researchers' experience. In most cases, the segmentation process and the travel demand modeling are estimated separately in a two-stage process, and therefore there is no feedback between them. The experience of involving segmentation methodology in travel demand models is quite basic.
The current research proposes the development of an integrated methodological framework, the Flexible Model Structure (FMS), to enhance the application of a Discrete Choice model by developing an optimization algorithm, which segments given data and searches for the best model structure for each segment. The process, searches for the best model structure match for each group.
The segmentation process relies on the fuzzy method. Moreover, the variables used in the segmentation process are treated with different weights to determine the belonging of an individual to a particular segment. The model structure search relies on the same concept, on the fuzzy method.
The FMS and some of its special cases were estimated in two different case studies using two different travel habit survey data sources from Israel.
The estimation results support the research hypotheses and emphasize the necessity of simultaneous estimation of the segmentation process and model structure search to better understand travel behavior. The main improvement in the estimation results is derived from the segmentation processes, which is innovative for this research.
The implementation of the FMS, which achieved the best estimation results, shows the travel behavior of an individual as a mix of travel behaviors represented in a number of segments, and that the choice model of each segment is a combination of different choice model structures.
Besides the FMS, which achieved the best estimation result, the implementation of Latent Class Model (LCM) using fuzzy segmentation methodology, as a special case of FMS, showed a great improvement in the estimation results, compared with the Mixed Logit Model.